you can generate the tfrecords by running this: python setup_updated/gen_tfr_multi2_mask.py once tfrecords are generated you can split the text files into train and val by running this: python train_test_split.py